CN112345240B - Mechanical part fault diagnosis system - Google Patents
Mechanical part fault diagnosis system Download PDFInfo
- Publication number
- CN112345240B CN112345240B CN201910727965.3A CN201910727965A CN112345240B CN 112345240 B CN112345240 B CN 112345240B CN 201910727965 A CN201910727965 A CN 201910727965A CN 112345240 B CN112345240 B CN 112345240B
- Authority
- CN
- China
- Prior art keywords
- signal
- detection signals
- detection
- sample
- signals
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M13/00—Testing of machine parts
- G01M13/04—Bearings
- G01M13/045—Acoustic or vibration analysis
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M13/00—Testing of machine parts
- G01M13/04—Bearings
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Acoustics & Sound (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention discloses a mechanical part fault diagnosis system.A state identification matrix construction module is used for obtaining an identification matrix for pattern identification of a mechanical part detection signal; the state identification matrix construction module calculates the significance of the factors by adopting a variance analysis method, and reconstructs a final state identification matrix for identifying the detection signal pattern to be distinguished based on the significance of the factors, so that the influence of non-significant factors can be eliminated, the accuracy of subsequent pattern identification is improved, the calculation is simpler and more convenient, and the realization is convenient.
Description
Technical Field
The invention relates to an automatic detection technology, in particular to a mechanical part fault diagnosis system.
Background
In the fields of industrial production and daily life, the diagnosis of wear or damage of mechanical parts is an important issue concerning life and production safety. In order to ensure the normal operation of equipment, a regular maintenance mode is usually adopted at present, but for the condition of low failure rate, the regular maintenance increases the cost, and for the condition of high failure rate, insufficient maintenance is likely to occur, and serious economic loss is caused in some occasions. Under the background, the fault diagnosis technology is developed, the running states of the parts of the equipment can be analyzed under the condition of no shutdown through monitoring equipment signals, and the risk and the maintenance cost in the running process are reduced.
In the actual operation process, monitoring signals of equipment parts often present obvious nonlinear non-stationary characteristics and are difficult to be directly used for analysis, the monitoring signals need to be decomposed into a plurality of stationary signals, component numbers may be different after different signals are decomposed, in this case, the numbers of signal characteristics extracted by different signals are different, and thus the signal characteristics cannot be used for fault mode identification. The Chinese patent granted Specification CN102854015B discloses a solution for solving the problem that the number of components is different after signal decomposition: firstly, the maximum number n of components after different signal decompositions is determined, and the number of the components is ensured to be consistent by the supplementary zero vector less than n. But due to the addition of the irrelevant elements, not only the efficiency of subsequent fault diagnosis is reduced, but also more serious problems are that: because the same zero element vector is supplemented, the difference between samples is reduced, and the misjudgment rate can be increased in the process of identifying the signal pattern.
The extraction of the signal characteristics is the premise of fault mode identification, but certain signal characteristics are not the key factors of fault mode identification, and non-key factors in the signal characteristics are removed and used for fault mode identification, so that the fault mode identification efficiency and the diagnosis accuracy can be improved. Chinese patent granted specification CN103674511B discloses that signal characteristics are reduced by a minimalist method, wherein a signal-to-noise ratio is used as an index to determine whether a certain signal characteristic should be removed, but the signal-to-noise ratio of an actual signal is difficult to obtain, noise variance and useful signal energy need to be estimated, the error of the estimated result is high, if the estimation accuracy is to be improved, a large number of samples are required for estimation, but a large number of faulty samples are often difficult to obtain, and at the same time, the workload of signal acquisition is greatly increased.
Based on the characteristic information of the signal, pattern recognition can be performed to judge the health degree of the mechanical parts behind the signal. The K-nearest neighbor (KNN) method is a nonparametric identification method, does not need to determine the prior probability and the quasi-conditional probability density function of a sample in advance, and therefore does not need a large number of samples for training. The Chinese patent granted Specification CN103488561B adopts a KNN method to detect faults. However, the selection of the neighbor number K in the KNN method has subjectivity, and different K value selections may result in different pattern recognition results, which affects the accuracy of judgment; and meanwhile, the difference of loss caused by misjudgment is not considered.
Disclosure of Invention
The invention aims to provide a mechanical part fault diagnosis system, which can improve the accuracy of mode identification and is convenient to realize.
In order to solve the technical problem, the mechanical part fault diagnosis system provided by the invention comprises a state identification matrix construction module;
the state identification matrix construction module is used for obtaining an identification matrix for pattern recognition of mechanical part detection signals;
the state identification matrix construction module comprises the following working processes:
firstly, constructing a characteristic matrix of e sample detection signals; e is a positive integer greater than 1, the e sample detection signals include at least one normal sample detection signal and at least one faultA sample detection signal; a characteristic matrix A of the sample detection signal mn The (i, j) element a of ij An ith time domain amplitude value of a jth stationary signal component representing a stationary signal component of the sample detection signal, m being a positive integer, i being an integer from 1 to m, j being an integer from 1 to n, n being the number of components of the detection signal for analysis of such stationary signal component;
then, singular value decomposition or generalized eigenvalue calculation is carried out on the eigen matrices of the e sample detection signals respectively, singular value vectors or generalized eigenvalue vectors of the eigen matrices of the e sample detection signals are obtained respectively, and the singular value vectors or generalized eigenvalue vectors of the e sample detection signals form an initial state identification matrix;
after an initial state identification matrix is formed, the extracted singular value vector or generalized eigenvalue vector is taken as a factor, different detection signal types are taken as levels, a level mean value is determined, variance analysis is carried out, F values of all factors are calculated, and an F table is searched according to a given significance level alpha to obtain F α If F is less than or equal to F α If the corresponding factor has no significant influence on the pattern recognition, deleting the corresponding factor of the initial state identification matrix; otherwise, the corresponding factors have obvious influence on the pattern recognition, the corresponding factors of the initial state recognition matrix are reserved, the different factors are analyzed respectively, and finally the final state recognition matrix for the pattern recognition is obtained.
Preferably, the first and second liquid crystal display panels are,
SSB is a one-factor squared intragroup dispersion sum, SSE is a one-factor squared intragroup dispersion sum, DFB is a one-factor intragroup degree of freedom, and DFE is a one-factor intragroup degree of freedom.
Preferably, the mechanical part fault diagnosis system further comprises a pattern recognition module;
the pattern recognition module is used for carrying out pattern recognition on the detection signal to be distinguished according to the final state recognition matrix;
the mode identification module works as follows:
step (ii) of1. Determining the number K of samples for pattern recognition in the u-th signal in the e sample detection signals according to the factor number and the factor dispersion degree of each type of signal in the final state identification matrix u When the maximum value is taken by the following formula, the w value is the corresponding K u The detection signals are classified into P types, P is an integer larger than 1, and u is an integer from 1 to P;
wherein n is u Number of samples of detection signal for u-th class, V (t) u /n u ) Is centered at t u The volume of the hyper-sphere of (a); t is t u Is n u Spatial center of a u-th detection signal u ；
Step two, respectively searching K closest to each sample detection signal in each type of e sample detection signals based on the extracted singular value vector or generalized eigenvalue vector u And (3) each neighbor, taking the distance r of the farthest sample detection signal in the neighbor as a radius, taking the detection signal to be distinguished as a center, calculating the volume of the corresponding hyper-sphere, and calculating the space ratio of the neighbor samples in the hyper-sphere volume according to the volume:
s is a detection signal to be distinguished; v (s/n) u ) A hypersphere volume centered at s;
step three, giving the misjudgment loss C (y | x) for misjudging the x-th class sample into the y-th class, and calculating the possible misjudgment loss G when the detection signal to be judged is classified into the z-th class z X, y and z are integers from 1 to P, and the expression is
Each obtained possible misjudgment loss G z Comparing the signals to classify the detected signals sTo the class with the least misjudgment loss.
Preferably, in step three, a misjudgment loss C (y | x) for misjudging the x-th type sample as the y-th type sample is given, wherein the misjudgment loss for misjudging the fault sample as a normal misjudgment loss is not less than the misjudgment loss for misjudging the normal sample as a fault.
Preferably, the mechanical part fault diagnosis system further comprises a fault detection sensor and a signal component number determination module;
the fault detection sensor is used for acquiring detection signals of mechanical parts;
the signal component number determining module is used for decomposing the h detection signals to obtain a stationary signal component of the detection signals, and then obtaining the number n of the components for analyzing the stationary signal component of the detection signals, wherein h is a positive integer;
the determination method of n is as follows: respectively calculating the ratio of the index of each stationary signal component of the s-th detection signal to the total index of the s-th detection signal, namely the index ratio, sequentially superposing the index ratios of the stationary signal components of the s-th detection signal according to the index ratios, stopping superposition when the sum of the index ratios is greater than or equal to the set ratio, and recording the number Ys of the stationary signal components of the s-th detection signal participating in superposition at the moment, wherein s is an integer from 1 to h; taking the maximum value Y of Y1 to Yh max Taking the minimum value of the number of the steady signal components of the h detection signals as X min Taking Y max And X min The relatively small value of both is n.
Preferably, the mechanical part fault diagnosis system further comprises a noise reduction processing module;
the noise reduction processing module is used for carrying out noise reduction processing on the mechanical part detection signals acquired by the fault detection sensor and then sending the signals to the signal component number determining module; the signal component number determining module decomposes the de-noised detection signal to obtain the number n of components for analyzing one stationary signal component of the detection signal.
Preferably, the detection signals of the mechanical parts are classified into normal and fault.
Preferably, the h detection signals are normal detection signals; alternatively, the first and second liquid crystal display panels may be,
the h detection signals are fault detection signals; alternatively, the first and second electrodes may be,
the h detection signals are normal detection signals and fault detection signals.
Preferably, the fault detection signals are classified into medium wear and heavy wear.
Preferably, the setting accounts for 90% to 100%.
Preferably, the mechanical part fault is a bearing wear fault;
the fault detection sensor adopts an acceleration sensor;
the acceleration sensor is used for collecting vibration acceleration signals of the bearing.
Preferably, the acceleration sensor is installed on the side surface of the bearing seat.
Preferably, the fault detection sensor is a non-vibration sensor.
Preferably, the non-vibration sensor is an acoustic sensor or an eddy current sensor
Preferably, the signal component number determining module is configured to perform inherent time scale decomposition ITD on the h detection signals to obtain a rotation component PRC as a stationary signal component, and then obtain a component number n for analysis of the rotation component of the detection signal;
the indicator is energy.
Preferably, the signal component number determining module is configured to perform empirical mode decomposition EMD on the h detection signals to obtain an intrinsic mode function component IMF as a stationary signal component, and then obtain an analysis component number n of the intrinsic mode function component IMF of the detection signals.
Preferably, the signal component number determining module is configured to perform ensemble empirical mode decomposition EEMD on the h detection signals to obtain an intrinsic mode function component IMF as a stationary signal component, and then obtain a component number n for analysis of the intrinsic mode function component IMF of the detection signal.
According to the mechanical part fault diagnosis system, the state identification matrix construction module adopts a variance analysis method to calculate the factor significance, and a final state identification matrix used for identifying the detection signal pattern to be distinguished is reconstructed on the basis of the factor significance, so that the influence of non-significant factors can be eliminated, the accuracy of subsequent pattern identification is improved, the calculation is simpler and more convenient, and the realization is convenient.
Drawings
In order to more clearly illustrate the technical solution of the present invention, the drawings required for the present invention are briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained according to the drawings without creative efforts.
FIG. 1 is a schematic structural diagram of an embodiment of a mechanical component fault diagnosis system of the present invention;
FIG. 2 is a schematic diagram of a working process of a state identification matrix building module according to an embodiment of the mechanical component fault diagnosis system of the present invention;
FIG. 3 is a schematic diagram of an embodiment of a pattern recognition module of the mechanical component fault diagnosis system according to the present invention;
FIG. 4 is a schematic diagram of a working process of a signal component number determination module according to an embodiment of the mechanical component fault diagnosis system of the present invention;
FIG. 5 is a diagram of an amplitude distribution of a detection signal to be discriminated in a time domain;
fig. 6 is a diagram illustrating an example of pattern recognition by the neighbor method.
Detailed Description
The technical solutions in the present invention will be described clearly and completely with reference to the accompanying drawings, and it should be understood that the described embodiments are some, but not all embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Example one
As shown in fig. 1, the mechanical component fault diagnosis system includes a state identification matrix building module;
the state identification matrix construction module is used for obtaining an identification matrix for pattern recognition of mechanical part detection signals;
the state identification matrix construction module, as shown in fig. 2, works as follows:
firstly, constructing a feature matrix of e sample detection signals; e is a positive integer greater than 1, the e sample detection signals include at least one normal sample detection signal and at least one fault sample detection signal; a characteristic matrix A of the sample detection signal mn The (i, j) element a of ij An ith time domain amplitude value of a jth stationary signal component representing a stationary signal component of the sample detection signal, m being a positive integer, i being an integer from 1 to m, j being an integer from 1 to n, n being the number of components of the detection signal for analysis of such stationary signal component;
then, singular value decomposition or generalized eigenvalue calculation is carried out on the eigenvalues of the e sample detection signals respectively, singular value vectors or generalized eigenvalue vectors of the eigenvalues of the e sample detection signals are obtained respectively, and the singular value vectors or generalized eigenvalue vectors of the e sample detection signals form an initial state identification matrix;
after an initial state identification matrix is formed, the extracted singular value vector or generalized eigenvalue vector is taken as a factor, different detection signal types are taken as levels, a level mean value is determined, variance analysis is carried out, F values of all factors are calculated, and an F table is searched according to a given significance level alpha to obtain F α If F is less than or equal to F α If F is less than or equal to F α If the corresponding factor has no significant influence on the pattern recognition, deleting the corresponding factor of the initial state identification matrix; otherwise, the corresponding factors have obvious influence on pattern recognition, the corresponding factors of the initial state recognition matrix are reserved, the different factors are analyzed respectively, and finally the final state recognition matrix for pattern recognition is obtained.
SSB is the square sum of inter-group dispersion of a factor, SSE is the square sum of intra-group dispersion of a factor, DFB is the inter-group degree of freedom of a factor, DFE is the intra-group degree of freedom of a factor.
According to the mechanical part fault diagnosis system, the state identification matrix construction module adopts a variance analysis method to calculate the factor significance, and the final state identification matrix for identifying the detection signal pattern to be distinguished is reconstructed on the basis of the factor significance, so that the influence of non-significant factors can be eliminated, the accuracy of subsequent pattern identification is improved, the calculation is simpler and more convenient, and the realization is convenient.
Example two
Based on the first embodiment, the mechanical part fault diagnosis system further comprises a mode identification module;
the pattern recognition module is used for carrying out pattern recognition on the detection signal to be distinguished according to the final state recognition matrix;
the mode identification module, as shown in fig. 3, works as follows:
step one, determining the number K of samples for pattern recognition in the u-th signal in the e sample detection signals according to the factor number and the factor dispersion degree of each type of signals in the final state identification matrix u When the maximum value is taken by the following formula, the w value is the corresponding K u The detection signals are divided into P types, P is an integer larger than 1, and u is an integer from 1 to P;
wherein n is u Number of samples of the detection signal for the u-th class, V (t) u /n u ) Is centered at t u The volume of the hyper-sphere of (a); t is t u Is n u Spatial center of a u-th detection signal u ；
Step two, searching K closest to each other in various sample detection signals in the e sample detection signals based on the extracted singular value vector or generalized eigenvalue vector u One neighbor, detecting the distance of the signal with the farthest sample in the neighborTaking the distance r as the radius, taking the detection signal to be distinguished as the center, calculating the corresponding hypersphere volume, and accordingly calculating the space ratio of the adjacent samples in the hypersphere volume:
s is a detection signal to be distinguished; v (s/n) u ) A hypersphere volume centered at s;
step three, giving the misjudgment loss C (y | x) for misjudging the x-th class sample into the y-th class, and calculating the possible misjudgment loss G when the detection signal to be judged is classified into the z-th class z X, y and z are integers from 1 to P, and the expression is
Each obtained possible misjudgment loss G z And comparing, and classifying the detection signal s to be judged into the class with the minimum misjudgment loss.
Preferably, in step three, a misjudgment loss C (y | x) for misjudging the x-th type sample as the y-th type sample is given, wherein the misjudgment loss for misjudging the fault sample as a normal misjudgment loss is not less than the misjudgment loss for misjudging the normal sample as a fault. This makes the fault diagnosis more safe even if misjudgment occurs.
In the mechanical part fault diagnosis system according to the second embodiment, the pattern recognition module can perform pattern recognition on the signals to be detected according to the final state identification matrix, and in the pattern recognition process, the number of factors and the dispersion degree of the factors of various signals are considered, so that the number K of samples for pattern recognition in various signals is determined, the influence of the subjectivity of K value selection on the fault discrimination accuracy is avoided, the misjudgment loss is considered, a large number of samples are not required for training, the loss caused by the misjudgment is reduced, and the system is more suitable for health monitoring of actual equipment.
EXAMPLE III
Based on the first embodiment, the mechanical part fault diagnosis system comprises a fault detection sensor and a signal component number determination module;
the fault detection sensor is used for acquiring detection signals of mechanical parts; each detection signal has at least 2 stationary signal components;
the signal component number determining module is used for decomposing the h detection signals to obtain a stationary signal component of the detection signals, and then obtaining the number n of the components for analyzing the stationary signal component of the detection signals, wherein h is a positive integer;
as shown in fig. 4, n is determined in the following manner: respectively calculating the ratio of the index of each stationary signal component of the s-th detection signal to the total index of the s-th detection signal, namely the index ratio, sequentially superposing the index ratios of the stationary signal components of the s-th detection signal according to the index ratios, stopping superposition when the sum of the index ratios is greater than or equal to the set ratio, and recording the number Ys of the stationary signal components of the s-th detection signal participating in superposition at the moment, wherein s is an integer from 1 to h; taking the maximum value Y of Y1 to Yh max Taking the minimum value of the number of the stationary signal components of the h detection signals as X min Taking Y max And X min The relatively small value of both is n.
The mechanical part fault diagnosis system of the third embodiment calculates a superimposable index of a steady signal component of a detection signal, wherein the superimposable index has definite physical significance and can represent partial characteristics of the detection signal, and simultaneously satisfies the condition that the sum of the indexes of the steady signal components of the detection signal is less than or equal to the index of the detection signal, and then calculates the index proportion of the steady signal component; and obtaining the number n of the components for analyzing the stationary signal components of the detection signals according to the number of the stationary signal components of each detection signal participating in superposition. The mechanical part fault diagnosis system determines the number n of components for analyzing the stable signal components of a detection signal by utilizing the index ratio of the stable signal components, and quantitatively gives a signal component selection principle, so that the number of the signal components for characteristic extraction is consistent, the problem that the number of the signal components of various different signals is obviously different under the same set value is avoided, the signal components can be used for fault mode identification, elements irrelevant to the signals do not need to be added, the subsequent fault diagnosis efficiency cannot be reduced, the difference among samples cannot be reduced, and the misjudgment rate in the process of signal mode identification is reduced.
Example four
Based on the third embodiment, the mechanical part fault diagnosis system further comprises a noise reduction processing module;
the noise reduction processing module is used for carrying out noise reduction processing on the mechanical part detection signals acquired by the fault detection sensor and then sending the signals to the signal component number determining module; the signal component number determining module decomposes the de-noised detection signal to obtain the number n of components for analyzing a stationary signal component of the detection signal.
Preferably, the detection signals of the mechanical parts are classified into normal and fault.
Preferably, the h detection signals are normal detection signals; alternatively, the first and second liquid crystal display panels may be,
the h detection signals are fault detection signals; alternatively, the first and second liquid crystal display panels may be,
the h detection signals are normal detection signals and fault detection signals.
Preferably, the fault detection signal is divided into a medium wear type and a heavy wear type.
Preferably, the set proportion is 90% to 100%. When the set ratio is 100%, each stable signal component of the detection signal participates in superposition at the moment so as to meet the requirement of the set ratio.
EXAMPLE five
Mechanical part fault diagnosis system based on the third embodiment, wherein the mechanical part fault is a bearing wear fault
The fault detection sensor adopts an acceleration sensor;
the acceleration sensor is used for acquiring vibration acceleration signals of the bearing.
Preferably, the acceleration sensor is installed on the side surface of the bearing seat.
Example six
Based on the mechanical part fault diagnosis system of the third embodiment, the fault detection sensor adopts a non-vibration sensor;
preferably, the non-vibrating sensor is an acoustic sensor or an eddy current sensor.
EXAMPLE seven
Based on the mechanical part fault diagnosis system of the third embodiment, the signal component number determining module is used for carrying out inherent time scale decomposition (ITD) on the h detection signals to obtain a rotation component PRC as a stable signal component, and then obtaining the number n of components for analysis of the rotation component of the detection signals;
the indicator is energy.
Example eight
Based on the mechanical part fault diagnosis system of the third embodiment, the signal component number determination module is configured to perform Empirical Mode Decomposition (EMD) on the h detection signals to obtain an intrinsic mode function component (IMF) as a stationary signal component, and then obtain an analysis component number n of the intrinsic mode function component (IMF) of the detection signals.
Example nine
Based on the mechanical part fault diagnosis system of the third embodiment, the signal component number determination module is configured to perform ensemble empirical mode decomposition EEMD on the h detection signals, obtain an intrinsic mode function component IMF as a stationary signal component, and then obtain an analysis component number n of the intrinsic mode function component IMF of the detection signals.
Example ten
Mechanical part fault diagnosis system based on the third embodiment, wherein the mechanical part fault is a bearing wear fault
The fault detection sensor adopts an acceleration sensor;
the acceleration sensor is used for collecting vibration acceleration signals of the bearing.
The bearing vibration acceleration signals are divided into three types of normal, moderate abrasion and severe abrasion, the number of the u-th type detection signals is m, and each u-th type detection signal has at least 2 stable signal components;
the 2 nd detection signal is moderate abrasion, and m = 6;
the 3 rd detection signal is serious abrasion, and m = 6;
inherent time scale decomposition (ITD) is performed on 6 type 1 detection signals (normal detection signals), 6 type 2 detection signals (moderate wear detection signals), and 6 type 3 detection signals (severe wear detection signals), to obtain a rotation component (PRC) as a stationary signal component, and the number of PRC components of each detection signal is shown in table 1;
table 1: number of PRC components of each detection signal
The number determining module of the signal components decomposes the 1 st type detection signal (normal detection signal) to obtain the number n of the components for analyzing the stationary signal components of the detection signal;
the determination method of n is as follows: respectively calculating the ratio of the energy of each rotation component (PRC) of the s-th detection signal to the total energy of the s-th detection signal, namely an index ratio, sequentially superposing the index ratios of the rotation components (PRC) of the s-th detection signal according to the index ratios, stopping superposition when the superposition value of the index ratios is greater than or equal to a set ratio, recording the number Ys of the rotation components (PRC) of the s-th detection signal participating in superposition at the moment, wherein s is an integer from 1 to h; taking the maximum value Y of Y1 to Y6 max Taking the minimum value of the number of the rotation components (PRC) of the h detection signals as X min Taking Y max And X min The relatively small value of both is n.
TABLE 2 energy of Normal detection Signal rotation component (PRC)The ratio of the quantity to the total energy of the signal is set to be 95%, and the number of the rotation components (PRC) participating in superposition of the 1 st normal detection signal (normal 1) and the 4 th normal detection signal (normal 4) is 4; the number of the rotation components (PRC) which participate in the superposition of the 2 nd normal detection signal (normal 2), the 3 rd normal detection signal (normal 3), the 5 th normal detection signal (normal 5) and the 6 th normal detection signal (normal 6) is 5, and thus it is known that the maximum value Y among Y1 to Y6 is Y max Taking 5, the minimum value 7 of the total number of the rotation components (PRC) of each type 1 detection signal (normal detection signal) smaller than the sample population, and the signal component number determination module thereby determines that the number of the rotation components (PRC) for analysis of the type 1 detection signal (normal detection signal) is n =5;
table 2: ratio of energy of normal detection signal rotation component (PRC) to total signal energy
After determining that the number of analysis rotation components PRC of the class 1 detection signal (normal detection signal) is n =5; and (3) constructing a characteristic matrix of e sample detection signals, selecting 5 rotation components PRC for analysis for each sample detection signal, performing singular value decomposition on the matrix, and extracting signal characteristics.
In addition, vibration signals s of a group of bearings are collected, the time domain distribution of the signals is shown in fig. 5, PRC components are obtained through ITD decomposition after noise reduction, 5 PRC components are selected to form a characteristic matrix, and singular value decomposition is carried out to obtain singular values.
The traditional KNN nearest neighbor method is adopted for analysis, the K value is 3,1 of 3 sample signals closest to the K value is a serious abrasion signal, and 2 of the 3 sample signals are moderate abrasion signals, so that the signal to be classified belongs to the medium abrasion class.
EXAMPLE eleven
By adopting 18 groups of signals in the tenth embodiment, empirical mode decomposition EMD is performed on normal signals and fault signals, 5 intrinsic mode function components IMF are selected for each sample detection signal from intrinsic mode function components IMF obtained by decomposing each group of signals to form a feature matrix, singular value decomposition is performed on the feature matrix, singular value vectors are determined according to the obtained diagonal matrix, and specific results are shown in table 3.
Table 3: singular value of signal
The mean values of the levels in table 3 were determined using the singular values as factors and the different classes as levels.
The variance-squared sums including the intergroup variance-squared sum SSB and the intracoup variance-squared sum SSE were calculated, and the results corresponding to 5 factors are shown in table 4.
Table 4: sum of squared deviations SSB between groups, sum of squared deviations SSE within groups results
Dividing the sum of squared deviations by the respective degrees of freedom to obtain corresponding mean square deviations, and setting the inter-group degree of freedom as DFB and the intra-group degree of freedom as DFE, and calculating the mean square deviation MSB and the mean square deviation MSE of the inter-group as MSE
F values were calculated from the interclass mean square error MSB and the intraclass mean square error MSE, F = MSB/MSE. The degrees of freedom and the F value calculation results are shown in table 5.
Table 5: degree of freedom and F value results
Here, given a significance level α =0.01, the F distribution table was used to find the critical value F α Comparing F and F α The results show that the different levels of the first four factors have significant influence, the first four factors should be kept, and the F value of the last factor is smaller than F α The last factor needs to be deleted to obtain a state recognition matrix for pattern recognition.
In addition, vibration signals s of a group of bearings are collected, the time domain distribution of the signals is shown in fig. 5, IMF components are obtained through EMD after noise reduction, 5 IMF components are selected to form a characteristic matrix, and singular value decomposition is carried out to obtain singular values.
The traditional KNN nearest neighbor method is adopted for analysis, the K value is 3, 2 of 3 sample signals which are closest to the K value are serious wear signals, and 1 sample signal is a medium wear signal, so that the signal to be classified belongs to the serious wear class.
EXAMPLE twelve
By adopting 18 groups of signals in the tenth embodiment, performing Ensemble Empirical Mode Decomposition (EEMD) on normal signals and fault signals, decomposing each group of signals to obtain a plurality of intrinsic mode function components (IMFs), selecting energy as a superposable index, calculating the ratio of the energy of the IMF components of the normal signals to the total energy of the signals, superposing the energy ratios of the IMF components according to the descending order of the energy ratios, taking the set value as 95%, and determining that the number n =5 of analysis components of the intrinsic mode function components (IMFs) of the detection signals.
Constructing a feature matrix of e sample detection signals, selecting 5 intrinsic mode function components IMF for analysis for each sample detection signal, carrying out singular value decomposition on the feature matrix, giving significance level alpha =0.01, and finding out a critical value F by an F distribution table α Comparing F and F α The results show that the different levels of the first four factors have significant influence, the first four factors should be kept, and the F value of the last factor is smaller than F α And deletion is required, thereby obtaining a state recognition matrix for pattern recognition.
The pattern recognition module determines that the number of samples for pattern recognition in the normal signals is 2, the number of samples for pattern recognition in the medium wear signals is 2, and the number of samples for pattern recognition in the severe wear signals is 6 according to the number of samples and the dispersion degree of the various signals.
In addition, a group of vibration signals s of severely worn bearings are collected, the time domain distribution of the signals is shown in fig. 5, IMF components are obtained through EEMD decomposition after noise reduction, 4 IMF components are selected according to the descending order of the energy ratio to form a characteristic matrix, singular value decomposition is carried out to obtain singular values, and the singular value is shown in table 6.
Table 6: singular values of newly acquired signals
| |
|
Singular value 3 | Singular value 4 |
| 103.11946 | 72.48574 | 65.93745 | 60.39476 |
Searching 2 sample signals nearest to the characteristics of the signal to be distinguished in the characteristic space of the normal sample signals, taking the distance of the farthest sample in the sample signals as the radius, taking the signal to be distinguished as the center, and calculating the volume of the hyper-sphere, wherein the space ratio of the adjacent samples in the volume of the hyper-sphere is 1.61 multiplied by 10 -8 (ii) a Searching in feature space of moderate-wear sample signalsSearching 2 sample signals closest to the characteristics of the signal to be distinguished, taking the distance of the farthest sample in the sample signals as a radius and the signal to be distinguished as a center, and calculating the volume of the hypersphere, wherein the space ratio of adjacent samples in the volume of the hypersphere is 9.35 multiplied by 10 -8 (ii) a Searching 6 sample signals closest to the characteristics of the signals to be distinguished in the characteristic space of the severely worn sample signals, calculating the volume of the hypersphere by taking the distance of the farthest sample in the sample signals as the radius and the signals to be distinguished as the center, wherein the space ratio of adjacent samples in the hypersphere volume is 2.02 multiplied by 10 -7 。
For the sake of simplifying the calculation, it is assumed that the misjudgment losses are equal, and C (x | y) =1 is generally taken, where x is a column flag and y is a column flag, and C (x | y) can be represented by the following matrix
The calculation shows that the possible misjudgment losses of the newly acquired signals classified into normal, moderate abrasion and severe abrasion are respectively 2.96 multiplied by 10 -7 ，2.19×10 -7 And 1.10X 10 -7 And if so, classifying the bearing into a serious abrasion class, and consistent with the classification of the actual fault condition of the bearing.
EXAMPLE thirteen
The fault detection sensor adopts a non-vibration sensor (such as an acoustic sensor or an eddy current sensor) to acquire a detection signal of a mechanical part; the detection signal samples are classified into two types, a type and B type.
And after the detection signals are filtered, performing wavelet packet decomposition to obtain 2 analysis components of a stable signal component of the detection signals, constructing a characteristic matrix of e sample detection signals according to the 2 analysis components, performing singular value decomposition on the characteristic matrix, and extracting signal characteristics. The characteristic values of the class A samples are respectively as follows: (6, 18), (7, 26), (8, 41), (12, 26), (18, 19), (25, 38), and the characteristic values of the class B samples are: (25, 15), (28, 21), (19, 45), (30, 47) and (39, 41), the characteristic value of the signal to be classified is (23, 19), and the distribution of each sample in space is shown in fig. 5.
When the conventional KNN nearest neighbor method is used for analysis, if the K value is 3, the nearest 3 points are located in the range of the dotted circle in fig. 5, wherein the rectangles of the class B signals are more, and therefore the signal to be classified belongs to the class B signals.
If the value K is 5, the nearest 5 points are located in the range of the solid line circle in fig. 6, where the circle of the signal of class a is more, and the signal to be classified belongs to the signal of class a.
Therefore, the selection of the number K of neighbors in the KNN method is subjective, and different K value selections may result in different pattern recognition results, which affects the accuracy of judgment.
The pattern recognition method in the invention is adopted to determine that the number of samples for pattern recognition in the A-type signal is 4 and the number of samples for pattern recognition in the B-type signal is 2 according to the number of samples and the dispersion degree of each type of signal, thereby avoiding the subjectivity of K value selection.
Searching 4 sample signals closest to the characteristics of the signals to be distinguished in the characteristic space of the A-type sample signals, taking the distance of the farthest sample in the sample signals as a radius and the signals to be distinguished as a center, calculating the volume of a hypersphere, wherein the space ratio of adjacent samples in the hypersphere volume is 5.22 multiplied by 10 -4 (ii) a Similarly, 2 sample signals closest to the signal feature to be distinguished are searched in the feature space of the B-type sample signal, the distance of the farthest sample in the sample signals is taken as the radius, the signal to be distinguished is taken as the center, the hypersphere volume is calculated, and the space ratio of adjacent samples in the hypersphere volume is 2.20 multiplied by 10 -3 。
For the sake of simplifying the calculation, it is assumed that the misjudgment losses are equal, and C (x | y) =1 is generally taken, where x is a column flag and y is a column flag, and C (x | y) can be represented by the following matrix
Through calculation, the possible misjudgment loss of the signal to be judged classified into the A class and the B class is respectively 2.20 multiplied by 10 -3 And 5.22X 10 -4 And the classification result is B.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and should not be taken as limiting the scope of the present invention, and any modifications, equivalents, improvements and the like made within the spirit and principle of the present invention should be included in the scope of the present invention.
Claims (16)
1. A mechanical part fault diagnosis system is characterized by comprising a state identification matrix construction module;
the state identification matrix construction module is used for obtaining an identification matrix for pattern recognition of mechanical part detection signals;
the state identification matrix construction module comprises the following working processes:
firstly, constructing a characteristic matrix of e sample detection signals; e is a positive integer greater than 1, and the e sample detection signals include at least one normal sample detection signal and at least one fault sample detection signal; a characteristic matrix A of the sample detection signal mn Is (i, j) element a of ij An ith time domain amplitude value of a jth stationary signal component representing a stationary signal component of the sample detection signal, m being a positive integer, i being an integer from 1 to m, j being an integer from 1 to n, n being the number of components of the detection signal for analysis of such stationary signal component;
then, singular value decomposition or generalized eigenvalue calculation is carried out on the eigen matrices of the e sample detection signals respectively, singular value vectors or generalized eigenvalue vectors of the eigen matrices of the e sample detection signals are obtained respectively, and the singular value vectors or generalized eigenvalue vectors of the e sample detection signals form an initial state identification matrix;
after an initial state identification matrix is formed, the extracted singular value vector or generalized eigenvalue vector is taken as a factor, different detection signal types are taken as levels, a level mean value is determined, variance analysis is carried out, F values of all factors are calculated, and an F table is searched according to a given significance level alpha to obtain F α If F is less than or equal to F α If the corresponding factor has no significant influence on the pattern recognition, deleting the corresponding factor of the initial state identification matrix; otherwise, the corresponding factors have obvious influence on pattern recognition and the initial state is keptAnalyzing different factors respectively according to the corresponding factors of the state identification matrix, and finally obtaining a final state identification matrix for pattern recognition;
SSB is the square sum of inter-group dispersion of a factor, SSE is the square sum of intra-group dispersion of a factor, DFB is the inter-group degree of freedom of a factor, DFE is the intra-group degree of freedom of a factor.
2. The mechanical component failure diagnostic system of claim 1,
the mechanical part fault diagnosis system also comprises a mode identification module;
the pattern recognition module is used for carrying out pattern recognition on the detection signal to be distinguished according to the final state recognition matrix;
the mode identification module works as follows:
step one, determining the number K of samples for pattern recognition in the u-th signal in the e sample detection signals according to the factor number and the factor dispersion degree of each type of signals in the final state identification matrix u When the maximum value is taken by the following formula, the w value is the corresponding K u The detection signals are divided into P types, P is an integer larger than 1, and u is an integer from 1 to P;
wherein n is u Number of samples of detection signal for u-th class, V (t) u /n u ) Is centered at t u The volume of the hyper-sphere of (a); t is t u Is n u Spatial center of a u-th type detection signal u ；
Step two, searching K closest to each other in various sample detection signals in the e sample detection signals based on the extracted singular value vector or generalized eigenvalue vector u And (3) each adjacent neighbor, taking the distance r of the farthest sample detection signal in the adjacent neighbor as a radius, taking the detection signal to be judged as a center, calculating the volume of the corresponding hyper-sphere, and calculating the space ratio of the adjacent sample in the hyper-sphere volume according to the volume:
s is a detection signal to be distinguished; v (s/n) u ) A hypersphere volume centered at s;
step three, giving the misjudgment loss C (y | x) for misjudging the x-th class sample into the y-th class, and calculating the possible misjudgment loss G when the detection signal to be judged is classified into the z-th class z X, y and z are integers from 1 to P, and the expression is
Each obtained possible misjudgment loss G z And comparing, and classifying the detection signal s to be judged into the class with the minimum misjudgment loss.
3. The mechanical component failure diagnostic system of claim 2,
in the third step, the misjudgment loss C (y | x) for misjudging the x-th type sample into the y-th type sample is given, wherein the misjudgment loss for misjudging the fault sample into the normal type is not less than the misjudgment loss for misjudging the normal type sample into the fault.
4. The mechanical component failure diagnostic system of claim 1,
the mechanical part fault diagnosis system also comprises a fault detection sensor and a signal component number determination module;
the fault detection sensor is used for acquiring detection signals of mechanical parts;
the signal component number determining module is used for decomposing the h detection signals to obtain a stationary signal component of the detection signals, and then obtaining the number n of the components for analyzing the stationary signal component of the detection signals, wherein h is a positive integer;
the determination method of n is as follows: respectively calculating the ratio of the index of each stable signal component of the s-th detection signal to the total index of the s-th detection signal, namely the index ratio, sequentially superposing the index ratios of the stable signal components of the s-th detection signal according to the index ratios, stopping superposition when the superposition value of the index ratios is greater than or equal to the set ratio, recording the number Ys of the stable signal components of the s-th detection signal participating in superposition at the moment, wherein s is an integer from 1 to h; taking the maximum value Y of Y1 to Yh max Taking the minimum value of the number of the steady signal components of the h detection signals as X min Taking Y max And X min The relatively small value of both is n.
5. The mechanical component failure diagnostic system of claim 4,
the mechanical part fault diagnosis system also comprises a noise reduction processing module;
the noise reduction processing module is used for carrying out noise reduction processing on the mechanical part detection signals acquired by the fault detection sensor and then sending the signals to the signal component number determining module; the signal component number determining module decomposes the de-noised detection signal to obtain the number n of components for analyzing a stationary signal component of the detection signal.
6. The mechanical component failure diagnostic system of claim 4,
the detection signals of the mechanical parts are classified into normal and fault.
7. The mechanical component failure diagnostic system of claim 6,
the h detection signals are normal detection signals; alternatively, the first and second electrodes may be,
the h detection signals are fault detection signals; alternatively, the first and second electrodes may be,
the h detection signals are normal detection signals and fault detection signals.
8. The mechanical component failure diagnostic system of claim 6,
the fault detection signals are divided into two categories, medium wear and heavy wear.
9. The mechanical component failure diagnostic system of claim 4,
the setting takes 90% to 100% in percentage.
10. The mechanical component failure diagnostic system of claim 4,
the mechanical part fault is a bearing wear fault;
the fault detection sensor adopts an acceleration sensor;
the acceleration sensor is used for collecting vibration acceleration signals of the bearing.
11. The mechanical component failure diagnostic system of claim 10,
the acceleration sensor is arranged on the side surface of the bearing seat.
12. The mechanical component failure diagnostic system of claim 4,
the fault detection sensor is a non-vibration sensor.
13. The mechanical component failure diagnostic system of claim 12,
the non-vibrating sensor is an acoustic sensor or an eddy current sensor.
14. The mechanical component failure diagnostic system of claim 4,
the signal component number determining module is used for carrying out inherent time scale decomposition (ITD) on the h detection signals to obtain a rotation component PRC as a stable signal component, and then obtaining the number n of components for analysis of the rotation component of the detection signals;
the indicator is energy.
15. The mechanical component failure diagnostic system of claim 4,
the signal component number determining module is used for performing Empirical Mode Decomposition (EMD) on the h detection signals to obtain an intrinsic mode function component (IMF) as a stable signal component, and then obtaining a component number n for analysis of the intrinsic mode function component (IMF) of the detection signals.
16. The mechanical component failure diagnostic system of claim 4,
the signal component number determining module is used for carrying out Ensemble Empirical Mode Decomposition (EEMD) on the h detection signals to obtain an intrinsic mode function component (IMF) as a stable signal component, and then obtaining the component number n for analysis of the intrinsic mode function component (IMF) of the detection signals.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910727965.3A CN112345240B (en) | 2019-08-08 | 2019-08-08 | Mechanical part fault diagnosis system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910727965.3A CN112345240B (en) | 2019-08-08 | 2019-08-08 | Mechanical part fault diagnosis system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112345240A CN112345240A (en) | 2021-02-09 |
| CN112345240B true CN112345240B (en) | 2022-12-20 |
Family
ID=74367446
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910727965.3A Active CN112345240B (en) | 2019-08-08 | 2019-08-08 | Mechanical part fault diagnosis system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112345240B (en) |
Family Cites Families (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2005081635A2 (en) * | 2004-03-01 | 2005-09-09 | Yeda Research And Development Co. Ltd. | Sorting points into neighborhoods (spin) |
| CN101002682A (en) * | 2007-01-19 | 2007-07-25 | 哈尔滨工程大学 | Method for retrieval and matching of hand back vein characteristic used for identification of status |
| CN101799674B (en) * | 2010-03-28 | 2011-06-29 | 华中科技大学 | Method for analyzing service state of numerical control equipment |
| CN102867299B (en) * | 2012-08-09 | 2015-06-10 | 东华大学 | Image analysis method based on singular value decomposition and method applied to defect detection of fabric |
| CN103810382B (en) * | 2014-01-27 | 2017-01-25 | 中国电子科技集团公司第十研究所 | Method for choosing two-level data fusion strategy of airborne distributed multi-sensor |
| CN105490972B (en) * | 2014-09-26 | 2019-06-04 | 重庆邮电大学 | CDMA reverse data channel modulation identification method |
| CN105550426B (en) * | 2015-12-08 | 2018-08-28 | 东北大学 | A kind of multiple dimensioned binary tree blast furnace method for diagnosing faults based on sample decomposition |
| CN107563008A (en) * | 2017-08-08 | 2018-01-09 | 三峡大学 | Estimating operation reliability of tool method based on SVD conversion and supporting vector space |
-
2019
- 2019-08-08 CN CN201910727965.3A patent/CN112345240B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN112345240A (en) | 2021-02-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109710636B (en) | Unsupervised industrial system anomaly detection method based on deep transfer learning | |
| CN109816031B (en) | Transformer state evaluation clustering analysis method based on data imbalance measurement | |
| CN110263846A (en) | The method for diagnosing faults for being excavated and being learnt based on fault data depth | |
| CN102435910B (en) | Power electronic circuit health monitoring method based on support vector classification | |
| CN113962259B (en) | Multi-mode double-layer fault diagnosis method for fuel cell system | |
| CN107230013B (en) | Method for identifying abnormal power consumption and time positioning of distribution network users under unsupervised learning | |
| CN108665119B (en) | Water supply pipe network abnormal working condition early warning method | |
| CN111160776A (en) | Method for detecting abnormal working condition in sewage treatment process by utilizing block principal component analysis | |
| CN109858254A (en) | Platform of internet of things attack detection system and method based on log analysis | |
| CN113780443B (en) | Threat detection-oriented network security situation assessment method | |
| CN111273623A (en) | Fault diagnosis method based on Stacked LSTM | |
| CN113537321B (en) | Network flow anomaly detection method based on isolated forest and X mean value | |
| CN102507676A (en) | On-line drift compensation method of electronic nose based on multiple self-organizing neural networks | |
| CN115526515B (en) | Safety monitoring system of gate for water conservancy and hydropower | |
| CN109976308A (en) | A kind of extracting method of the fault signature based on Laplce's score value and AP cluster | |
| CN113918367A (en) | Large-scale system log anomaly detection method based on attention mechanism | |
| CN111382943A (en) | Fault diagnosis and evaluation method based on weighted grey correlation analysis | |
| CN114429152A (en) | Rolling bearing fault diagnosis method based on dynamic index antagonism self-adaption | |
| CN114462480A (en) | Multi-source sensor rolling mill fault diagnosis method based on non-equilibrium data set | |
| CN109434562A (en) | Milling cutter state of wear recognition methods based on partition clustering | |
| CN116208417A (en) | Communication abnormity sensing system and method based on big data | |
| CN112345240B (en) | Mechanical part fault diagnosis system | |
| CN109683594A (en) | A kind of exceptional variable accurately identifies and localization method | |
| CN112422546A (en) | Network anomaly detection method based on variable neighborhood algorithm and fuzzy clustering | |
| CN109767430B (en) | Quality detection method and quality detection system for valuable bills |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |